ViewVC Help

View File | Revision Log | Show Annotations | View Changeset | Root Listing

root/OpenMD/trunk/src/parallel/ForceMatrixDecomposition.cpp

Comparing branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents):
Revision 1581 by gezelter, Mon Jun 13 22:13:12 2011 UTC vs.
Revision 1761 by gezelter, Fri Jun 22 20:01:37 2012 UTC

#	Line 36 | Line 36
36		* [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).
37		* [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).
38		* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).
39	<	* [4]  Vardeman & Gezelter, in progress (2009).
39	>	* [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40	>	* [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41		*/
42		#include "parallel/ForceMatrixDecomposition.hpp"
43		#include "math/SquareMatrix3.hpp"
#	Line 47 | Line 48 | namespace OpenMD {
48		using namespace std;
49		namespace OpenMD {
50
51	+	ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
52	+
53	+	// In a parallel computation, row and colum scans must visit all
54	+	// surrounding cells (not just the 14 upper triangular blocks that
55	+	// are used when the processor can see all pairs)
56	+	#ifdef IS_MPI
57	+	cellOffsets_.clear();
58	+	cellOffsets_.push_back( Vector3i(-1,-1,-1) );
59	+	cellOffsets_.push_back( Vector3i( 0,-1,-1) );
60	+	cellOffsets_.push_back( Vector3i( 1,-1,-1) );
61	+	cellOffsets_.push_back( Vector3i(-1, 0,-1) );
62	+	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
63	+	cellOffsets_.push_back( Vector3i( 1, 0,-1) );
64	+	cellOffsets_.push_back( Vector3i(-1, 1,-1) );
65	+	cellOffsets_.push_back( Vector3i( 0, 1,-1) );
66	+	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67	+	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
68	+	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
69	+	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
70	+	cellOffsets_.push_back( Vector3i(-1, 0, 0) );
71	+	cellOffsets_.push_back( Vector3i( 0, 0, 0) );
72	+	cellOffsets_.push_back( Vector3i( 1, 0, 0) );
73	+	cellOffsets_.push_back( Vector3i(-1, 1, 0) );
74	+	cellOffsets_.push_back( Vector3i( 0, 1, 0) );
75	+	cellOffsets_.push_back( Vector3i( 1, 1, 0) );
76	+	cellOffsets_.push_back( Vector3i(-1,-1, 1) );
77	+	cellOffsets_.push_back( Vector3i( 0,-1, 1) );
78	+	cellOffsets_.push_back( Vector3i( 1,-1, 1) );
79	+	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
80	+	cellOffsets_.push_back( Vector3i( 0, 0, 1) );
81	+	cellOffsets_.push_back( Vector3i( 1, 0, 1) );
82	+	cellOffsets_.push_back( Vector3i(-1, 1, 1) );
83	+	cellOffsets_.push_back( Vector3i( 0, 1, 1) );
84	+	cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85	+	#endif
86	+	}
87	+
88	+
89		/**
90		* distributeInitialData is essentially a copy of the older fortran
91		* SimulationSetup
92		*/
54	–
93		void ForceMatrixDecomposition::distributeInitialData() {
94		snap_ = sman_->getCurrentSnapshot();
95		storageLayout_ = sman_->getStorageLayout();
96		ff_ = info_->getForceField();
97		nLocal_ = snap_->getNumberOfAtoms();
98	<
98	>
99		nGroups_ = info_->getNLocalCutoffGroups();
62	–	cerr << "in dId, nGroups = " << nGroups_ << "\n";
100		// gather the information for atomtype IDs (atids):
101	<	identsLocal = info_->getIdentArray();
101	>	idents = info_->getIdentArray();
102		AtomLocalToGlobal = info_->getGlobalAtomIndices();
103		cgLocalToGlobal = info_->getGlobalGroupIndices();
104		vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
105	+
106		massFactors = info_->getMassFactors();
69	–	PairList excludes = info_->getExcludedInteractions();
70	–	PairList oneTwo = info_->getOneTwoInteractions();
71	–	PairList oneThree = info_->getOneThreeInteractions();
72	–	PairList oneFour = info_->getOneFourInteractions();
107
108	+	PairList* excludes = info_->getExcludedInteractions();
109	+	PairList* oneTwo = info_->getOneTwoInteractions();
110	+	PairList* oneThree = info_->getOneThreeInteractions();
111	+	PairList* oneFour = info_->getOneFourInteractions();
112	+
113	+	if (needVelocities_)
114	+	snap_->cgData.setStorageLayout(DataStorage::dslPosition |
115	+	DataStorage::dslVelocity);
116	+	else
117	+	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
118	+
119		#ifdef IS_MPI
120
121	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
122	<	AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
78	<	AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
79	<	AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
80	<	AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
121	>	MPI::Intracomm row = rowComm.getComm();
122	>	MPI::Intracomm col = colComm.getComm();
123
124	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
125	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
126	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
127	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
128	<	AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
124	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
125	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
126	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
127	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
128	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
129
130	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
131	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
132	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
133	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
130	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
131	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
132	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
133	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
134	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
135
136	<	nAtomsInRow_ = AtomCommIntRow->getSize();
137	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
138	<	nGroupsInRow_ = cgCommIntRow->getSize();
139	<	nGroupsInCol_ = cgCommIntColumn->getSize();
136	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
137	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
138	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
139	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
140
141	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
142	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
143	+	nGroupsInRow_ = cgPlanIntRow->getSize();
144	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
145	+
146		// Modify the data storage objects with the correct layouts and sizes:
147		atomRowData.resize(nAtomsInRow_);
148		atomRowData.setStorageLayout(storageLayout_);
#	Line 103 | Line 151 | namespace OpenMD {
151		cgRowData.resize(nGroupsInRow_);
152		cgRowData.setStorageLayout(DataStorage::dslPosition);
153		cgColData.resize(nGroupsInCol_);
154	<	cgColData.setStorageLayout(DataStorage::dslPosition);
155	<
154	>	if (needVelocities_)
155	>	// we only need column velocities if we need them.
156	>	cgColData.setStorageLayout(DataStorage::dslPosition |
157	>	DataStorage::dslVelocity);
158	>	else
159	>	cgColData.setStorageLayout(DataStorage::dslPosition);
160	>
161		identsRow.resize(nAtomsInRow_);
162		identsCol.resize(nAtomsInCol_);
163
164	<	AtomCommIntRow->gather(identsLocal, identsRow);
165	<	AtomCommIntColumn->gather(identsLocal, identsCol);
164	>	AtomPlanIntRow->gather(idents, identsRow);
165	>	AtomPlanIntColumn->gather(idents, identsCol);
166
167	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
168	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
169	<
117	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
118	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
167	>	// allocate memory for the parallel objects
168	>	atypesRow.resize(nAtomsInRow_);
169	>	atypesCol.resize(nAtomsInCol_);
170
171	<	AtomCommRealRow->gather(massFactors, massFactorsRow);
172	<	AtomCommRealColumn->gather(massFactors, massFactorsCol);
171	>	for (int i = 0; i < nAtomsInRow_; i++)
172	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
173	>	for (int i = 0; i < nAtomsInCol_; i++)
174	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
175
176	+	pot_row.resize(nAtomsInRow_);
177	+	pot_col.resize(nAtomsInCol_);
178	+
179	+	expot_row.resize(nAtomsInRow_);
180	+	expot_col.resize(nAtomsInCol_);
181	+
182	+	AtomRowToGlobal.resize(nAtomsInRow_);
183	+	AtomColToGlobal.resize(nAtomsInCol_);
184	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
185	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
186	+
187	+	cgRowToGlobal.resize(nGroupsInRow_);
188	+	cgColToGlobal.resize(nGroupsInCol_);
189	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
190	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
191	+
192	+	massFactorsRow.resize(nAtomsInRow_);
193	+	massFactorsCol.resize(nAtomsInCol_);
194	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
195	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
196	+
197		groupListRow_.clear();
198		groupListRow_.resize(nGroupsInRow_);
199		for (int i = 0; i < nGroupsInRow_; i++) {
#	Line 142 | Line 216 | namespace OpenMD {
216		}
217		}
218
219	<	skipsForAtom.clear();
220	<	skipsForAtom.resize(nAtomsInRow_);
219	>	excludesForAtom.clear();
220	>	excludesForAtom.resize(nAtomsInRow_);
221		toposForAtom.clear();
222		toposForAtom.resize(nAtomsInRow_);
223		topoDist.clear();
#	Line 154 | Line 228 | namespace OpenMD {
228		for (int j = 0; j < nAtomsInCol_; j++) {
229		int jglob = AtomColToGlobal[j];
230
231	<	if (excludes.hasPair(iglob, jglob))
232	<	skipsForAtom[i].push_back(j);
231	>	if (excludes->hasPair(iglob, jglob))
232	>	excludesForAtom[i].push_back(j);
233
234	<	if (oneTwo.hasPair(iglob, jglob)) {
234	>	if (oneTwo->hasPair(iglob, jglob)) {
235		toposForAtom[i].push_back(j);
236		topoDist[i].push_back(1);
237		} else {
238	<	if (oneThree.hasPair(iglob, jglob)) {
238	>	if (oneThree->hasPair(iglob, jglob)) {
239		toposForAtom[i].push_back(j);
240		topoDist[i].push_back(2);
241		} else {
242	<	if (oneFour.hasPair(iglob, jglob)) {
242	>	if (oneFour->hasPair(iglob, jglob)) {
243		toposForAtom[i].push_back(j);
244		topoDist[i].push_back(3);
245		}
#	Line 174 | Line 248 | namespace OpenMD {
248		}
249		}
250
251	<	#endif
252	<
253	<	groupList_.clear();
180	<	groupList_.resize(nGroups_);
181	<	for (int i = 0; i < nGroups_; i++) {
182	<	int gid = cgLocalToGlobal[i];
183	<	for (int j = 0; j < nLocal_; j++) {
184	<	int aid = AtomLocalToGlobal[j];
185	<	if (globalGroupMembership[aid] == gid) {
186	<	groupList_[i].push_back(j);
187	<	}
188	<	}
189	<	}
190	<
191	<	skipsForAtom.clear();
192	<	skipsForAtom.resize(nLocal_);
251	>	#else
252	>	excludesForAtom.clear();
253	>	excludesForAtom.resize(nLocal_);
254		toposForAtom.clear();
255		toposForAtom.resize(nLocal_);
256		topoDist.clear();
#	Line 201 | Line 262 | namespace OpenMD {
262		for (int j = 0; j < nLocal_; j++) {
263		int jglob = AtomLocalToGlobal[j];
264
265	<	if (excludes.hasPair(iglob, jglob))
266	<	skipsForAtom[i].push_back(j);
265	>	if (excludes->hasPair(iglob, jglob))
266	>	excludesForAtom[i].push_back(j);
267
268	<	if (oneTwo.hasPair(iglob, jglob)) {
268	>	if (oneTwo->hasPair(iglob, jglob)) {
269		toposForAtom[i].push_back(j);
270		topoDist[i].push_back(1);
271		} else {
272	<	if (oneThree.hasPair(iglob, jglob)) {
272	>	if (oneThree->hasPair(iglob, jglob)) {
273		toposForAtom[i].push_back(j);
274		topoDist[i].push_back(2);
275		} else {
276	<	if (oneFour.hasPair(iglob, jglob)) {
276	>	if (oneFour->hasPair(iglob, jglob)) {
277		toposForAtom[i].push_back(j);
278		topoDist[i].push_back(3);
279		}
#	Line 220 | Line 281 | namespace OpenMD {
281		}
282		}
283		}
284	<
284	>	#endif
285	>
286	>	// allocate memory for the parallel objects
287	>	atypesLocal.resize(nLocal_);
288	>
289	>	for (int i = 0; i < nLocal_; i++)
290	>	atypesLocal[i] = ff_->getAtomType(idents[i]);
291	>
292	>	groupList_.clear();
293	>	groupList_.resize(nGroups_);
294	>	for (int i = 0; i < nGroups_; i++) {
295	>	int gid = cgLocalToGlobal[i];
296	>	for (int j = 0; j < nLocal_; j++) {
297	>	int aid = AtomLocalToGlobal[j];
298	>	if (globalGroupMembership[aid] == gid) {
299	>	groupList_[i].push_back(j);
300	>	}
301	>	}
302	>	}
303	>
304	>
305		createGtypeCutoffMap();
306	+
307		}
308
309		void ForceMatrixDecomposition::createGtypeCutoffMap() {
310	<
310	>
311		RealType tol = 1e-6;
312	+	largestRcut_ = 0.0;
313		RealType rc;
314		int atid;
315		set<AtomType*> atypes = info_->getSimulatedAtomTypes();
316	<	vector<RealType> atypeCutoff;
317	<	atypeCutoff.resize( atypes.size() );
318	<
316	>
317	>	map<int, RealType> atypeCutoff;
318	>
319		for (set<AtomType*>::iterator at = atypes.begin();
320		at != atypes.end(); ++at){
238	–	rc = interactionMan_->getSuggestedCutoffRadius(*at);
321		atid = (*at)->getIdent();
322	<	atypeCutoff[atid] = rc;
322	>	if (userChoseCutoff_)
323	>	atypeCutoff[atid] = userCutoff_;
324	>	else
325	>	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
326		}
327	<
327	>
328		vector<RealType> gTypeCutoffs;
244	–
329		// first we do a single loop over the cutoff groups to find the
330		// largest cutoff for any atypes present in this group.
331		#ifdef IS_MPI
#	Line 299 | Line 383 | namespace OpenMD {
383
384		vector<RealType> groupCutoff(nGroups_, 0.0);
385		groupToGtype.resize(nGroups_);
302	–
303	–	cerr << "nGroups = " << nGroups_ << "\n";
386		for (int cg1 = 0; cg1 < nGroups_; cg1++) {
305	–
387		groupCutoff[cg1] = 0.0;
388		vector<int> atomList = getAtomsInGroupRow(cg1);
308	–
389		for (vector<int>::iterator ia = atomList.begin();
390		ia != atomList.end(); ++ia) {
391		int atom1 = (*ia);
392	<	atid = identsLocal[atom1];
393	<	if (atypeCutoff[atid] > groupCutoff[cg1]) {
392	>	atid = idents[atom1];
393	>	if (atypeCutoff[atid] > groupCutoff[cg1])
394		groupCutoff[cg1] = atypeCutoff[atid];
315	–	}
395		}
396	<
396	>
397		bool gTypeFound = false;
398		for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
399		if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
#	Line 322 | Line 401 | namespace OpenMD {
401		gTypeFound = true;
402		}
403		}
404	<	if (!gTypeFound) {
404	>	if (!gTypeFound) {
405		gTypeCutoffs.push_back( groupCutoff[cg1] );
406		groupToGtype[cg1] = gTypeCutoffs.size() - 1;
407		}
408		}
409		#endif
410
332	–	cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n";
411		// Now we find the maximum group cutoff value present in the simulation
412
413	<	RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
413	>	RealType groupMax = *max_element(gTypeCutoffs.begin(),
414	>	gTypeCutoffs.end());
415
416		#ifdef IS_MPI
417	<	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
417	>	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
418	>	MPI::MAX);
419		#endif
420
421		RealType tradRcut = groupMax;
#	Line 365 | Line 445 | namespace OpenMD {
445
446		pair<int,int> key = make_pair(i,j);
447		gTypeCutoffMap[key].first = thisRcut;
368	–
448		if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
370	–
449		gTypeCutoffMap[key].second = thisRcut*thisRcut;
372	–
450		gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
374	–
451		// sanity check
452
453		if (userChoseCutoff_) {
454		if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
455		sprintf(painCave.errMsg,
456		"ForceMatrixDecomposition::createGtypeCutoffMap "
457	<	"user-specified rCut does not match computed group Cutoff\n");
457	>	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
458		painCave.severity = OPENMD_ERROR;
459		painCave.isFatal = 1;
460		simError();
#	Line 388 | Line 464 | namespace OpenMD {
464		}
465		}
466
391	–
467		groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
468		int i, j;
469		#ifdef IS_MPI
#	Line 410 | Line 485 | namespace OpenMD {
485		}
486
487		void ForceMatrixDecomposition::zeroWorkArrays() {
488	+	pairwisePot = 0.0;
489	+	embeddingPot = 0.0;
490	+	excludedPot = 0.0;
491	+	excludedSelfPot = 0.0;
492
414	–	for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
415	–	longRangePot_[j] = 0.0;
416	–	}
417	–
493		#ifdef IS_MPI
494		if (storageLayout_ & DataStorage::dslForce) {
495		fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
#	Line 430 | Line 505 | namespace OpenMD {
505		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
506
507		fill(pot_col.begin(), pot_col.end(),
508	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
509	+
510	+	fill(expot_row.begin(), expot_row.end(),
511		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
434	–
435	–	pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
512
513	+	fill(expot_col.begin(), expot_col.end(),
514	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
515	+
516		if (storageLayout_ & DataStorage::dslParticlePot) {
517	<	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
518	<	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
517	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
518	>	0.0);
519	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
520	>	0.0);
521		}
522
523		if (storageLayout_ & DataStorage::dslDensity) {
#	Line 445 | Line 526 | namespace OpenMD {
526		}
527
528		if (storageLayout_ & DataStorage::dslFunctional) {
529	<	fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
530	<	fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
529	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
530	>	0.0);
531	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
532	>	0.0);
533		}
534
535		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
#	Line 456 | Line 539 | namespace OpenMD {
539		atomColData.functionalDerivative.end(), 0.0);
540		}
541
542	<	#else
543	<
542	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
543	>	fill(atomRowData.skippedCharge.begin(),
544	>	atomRowData.skippedCharge.end(), 0.0);
545	>	fill(atomColData.skippedCharge.begin(),
546	>	atomColData.skippedCharge.end(), 0.0);
547	>	}
548	>
549	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
550	>	fill(atomRowData.flucQFrc.begin(),
551	>	atomRowData.flucQFrc.end(), 0.0);
552	>	fill(atomColData.flucQFrc.begin(),
553	>	atomColData.flucQFrc.end(), 0.0);
554	>	}
555	>
556	>	if (storageLayout_ & DataStorage::dslElectricField) {
557	>	fill(atomRowData.electricField.begin(),
558	>	atomRowData.electricField.end(), V3Zero);
559	>	fill(atomColData.electricField.begin(),
560	>	atomColData.electricField.end(), V3Zero);
561	>	}
562	>
563	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
564	>	fill(atomRowData.flucQFrc.begin(), atomRowData.flucQFrc.end(),
565	>	0.0);
566	>	fill(atomColData.flucQFrc.begin(), atomColData.flucQFrc.end(),
567	>	0.0);
568	>	}
569	>
570	>	#endif
571	>	// even in parallel, we need to zero out the local arrays:
572	>
573		if (storageLayout_ & DataStorage::dslParticlePot) {
574		fill(snap_->atomData.particlePot.begin(),
575		snap_->atomData.particlePot.end(), 0.0);
#	Line 467 | Line 579 | namespace OpenMD {
579		fill(snap_->atomData.density.begin(),
580		snap_->atomData.density.end(), 0.0);
581		}
582	+
583		if (storageLayout_ & DataStorage::dslFunctional) {
584		fill(snap_->atomData.functional.begin(),
585		snap_->atomData.functional.end(), 0.0);
586		}
587	+
588		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
589		fill(snap_->atomData.functionalDerivative.begin(),
590		snap_->atomData.functionalDerivative.end(), 0.0);
591		}
592	<	#endif
593	<
592	>
593	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
594	>	fill(snap_->atomData.skippedCharge.begin(),
595	>	snap_->atomData.skippedCharge.end(), 0.0);
596	>	}
597	>
598	>	if (storageLayout_ & DataStorage::dslElectricField) {
599	>	fill(snap_->atomData.electricField.begin(),
600	>	snap_->atomData.electricField.end(), V3Zero);
601	>	}
602		}
603
604
#	Line 486 | Line 608 | namespace OpenMD {
608		#ifdef IS_MPI
609
610		// gather up the atomic positions
611	<	AtomCommVectorRow->gather(snap_->atomData.position,
611	>	AtomPlanVectorRow->gather(snap_->atomData.position,
612		atomRowData.position);
613	<	AtomCommVectorColumn->gather(snap_->atomData.position,
613	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
614		atomColData.position);
615
616		// gather up the cutoff group positions
617	<	cgCommVectorRow->gather(snap_->cgData.position,
617	>
618	>	cgPlanVectorRow->gather(snap_->cgData.position,
619		cgRowData.position);
620	<	cgCommVectorColumn->gather(snap_->cgData.position,
620	>
621	>	cgPlanVectorColumn->gather(snap_->cgData.position,
622		cgColData.position);
623	+
624	+
625	+
626	+	if (needVelocities_) {
627	+	// gather up the atomic velocities
628	+	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
629	+	atomColData.velocity);
630	+
631	+	cgPlanVectorColumn->gather(snap_->cgData.velocity,
632	+	cgColData.velocity);
633	+	}
634	+
635
636		// if needed, gather the atomic rotation matrices
637		if (storageLayout_ & DataStorage::dslAmat) {
638	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
638	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
639		atomRowData.aMat);
640	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
640	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
641		atomColData.aMat);
642		}
643
644		// if needed, gather the atomic eletrostatic frames
645		if (storageLayout_ & DataStorage::dslElectroFrame) {
646	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
646	>	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
647		atomRowData.electroFrame);
648	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
648	>	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
649		atomColData.electroFrame);
650		}
651	+
652	+	// if needed, gather the atomic fluctuating charge values
653	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
654	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
655	+	atomRowData.flucQPos);
656	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
657	+	atomColData.flucQPos);
658	+	}
659	+
660		#endif
661		}
662
#	Line 525 | Line 670 | namespace OpenMD {
670
671		if (storageLayout_ & DataStorage::dslDensity) {
672
673	<	AtomCommRealRow->scatter(atomRowData.density,
673	>	AtomPlanRealRow->scatter(atomRowData.density,
674		snap_->atomData.density);
675
676		int n = snap_->atomData.density.size();
677		vector<RealType> rho_tmp(n, 0.0);
678	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
678	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
679		for (int i = 0; i < n; i++)
680		snap_->atomData.density[i] += rho_tmp[i];
681		}
682	+
683	+	if (storageLayout_ & DataStorage::dslElectricField) {
684	+
685	+	AtomPlanVectorRow->scatter(atomRowData.electricField,
686	+	snap_->atomData.electricField);
687	+
688	+	int n = snap_->atomData.electricField.size();
689	+	vector<Vector3d> field_tmp(n, V3Zero);
690	+	AtomPlanVectorColumn->scatter(atomColData.electricField, field_tmp);
691	+	for (int i = 0; i < n; i++)
692	+	snap_->atomData.electricField[i] += field_tmp[i];
693	+	}
694		#endif
695		}
696
#	Line 546 | Line 703 | namespace OpenMD {
703		storageLayout_ = sman_->getStorageLayout();
704		#ifdef IS_MPI
705		if (storageLayout_ & DataStorage::dslFunctional) {
706	<	AtomCommRealRow->gather(snap_->atomData.functional,
706	>	AtomPlanRealRow->gather(snap_->atomData.functional,
707		atomRowData.functional);
708	<	AtomCommRealColumn->gather(snap_->atomData.functional,
708	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
709		atomColData.functional);
710		}
711
712		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
713	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
713	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
714		atomRowData.functionalDerivative);
715	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
715	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
716		atomColData.functionalDerivative);
717		}
718		#endif
#	Line 569 | Line 726 | namespace OpenMD {
726		int n = snap_->atomData.force.size();
727		vector<Vector3d> frc_tmp(n, V3Zero);
728
729	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
729	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
730		for (int i = 0; i < n; i++) {
731		snap_->atomData.force[i] += frc_tmp[i];
732		frc_tmp[i] = 0.0;
733		}
734
735	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
736	<	for (int i = 0; i < n; i++)
735	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
736	>	for (int i = 0; i < n; i++) {
737		snap_->atomData.force[i] += frc_tmp[i];
738	<
739	<
738	>	}
739	>
740		if (storageLayout_ & DataStorage::dslTorque) {
741
742	<	int nt = snap_->atomData.force.size();
742	>	int nt = snap_->atomData.torque.size();
743		vector<Vector3d> trq_tmp(nt, V3Zero);
744
745	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
746	<	for (int i = 0; i < n; i++) {
745	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
746	>	for (int i = 0; i < nt; i++) {
747		snap_->atomData.torque[i] += trq_tmp[i];
748		trq_tmp[i] = 0.0;
749		}
750
751	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
752	<	for (int i = 0; i < n; i++)
751	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
752	>	for (int i = 0; i < nt; i++)
753		snap_->atomData.torque[i] += trq_tmp[i];
754		}
755	+
756	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
757	+
758	+	int ns = snap_->atomData.skippedCharge.size();
759	+	vector<RealType> skch_tmp(ns, 0.0);
760	+
761	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
762	+	for (int i = 0; i < ns; i++) {
763	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
764	+	skch_tmp[i] = 0.0;
765	+	}
766	+
767	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
768	+	for (int i = 0; i < ns; i++)
769	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
770	+
771	+	}
772
773	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
774	+
775	+	int nq = snap_->atomData.flucQFrc.size();
776	+	vector<RealType> fqfrc_tmp(nq, 0.0);
777	+
778	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
779	+	for (int i = 0; i < nq; i++) {
780	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
781	+	fqfrc_tmp[i] = 0.0;
782	+	}
783	+
784	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
785	+	for (int i = 0; i < nq; i++)
786	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
787	+
788	+	}
789	+
790		nLocal_ = snap_->getNumberOfAtoms();
791
792		vector<potVec> pot_temp(nLocal_,
793		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
794	+	vector<potVec> expot_temp(nLocal_,
795	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
796
797		// scatter/gather pot_row into the members of my column
798
799	<	AtomCommPotRow->scatter(pot_row, pot_temp);
799	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
800	>	AtomPlanPotRow->scatter(expot_row, expot_temp);
801
802	<	for (int ii = 0;  ii < pot_temp.size(); ii++ )
803	<	pot_local += pot_temp[ii];
804	<
802	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
803	>	pairwisePot += pot_temp[ii];
804	>
805	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
806	>	excludedPot += expot_temp[ii];
807	>
808	>	if (storageLayout_ & DataStorage::dslParticlePot) {
809	>	// This is the pairwise contribution to the particle pot.  The
810	>	// embedding contribution is added in each of the low level
811	>	// non-bonded routines.  In single processor, this is done in
812	>	// unpackInteractionData, not in collectData.
813	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
814	>	for (int i = 0; i < nLocal_; i++) {
815	>	// factor of two is because the total potential terms are divided
816	>	// by 2 in parallel due to row/ column scatter
817	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
818	>	}
819	>	}
820	>	}
821	>
822		fill(pot_temp.begin(), pot_temp.end(),
823		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
824	+	fill(expot_temp.begin(), expot_temp.end(),
825	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
826
827	<	AtomCommPotColumn->scatter(pot_col, pot_temp);
827	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
828	>	AtomPlanPotColumn->scatter(expot_col, expot_temp);
829
830		for (int ii = 0;  ii < pot_temp.size(); ii++ )
831	<	pot_local += pot_temp[ii];
831	>	pairwisePot += pot_temp[ii];
832	>
833	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
834	>	excludedPot += expot_temp[ii];
835	>
836	>	if (storageLayout_ & DataStorage::dslParticlePot) {
837	>	// This is the pairwise contribution to the particle pot.  The
838	>	// embedding contribution is added in each of the low level
839	>	// non-bonded routines.  In single processor, this is done in
840	>	// unpackInteractionData, not in collectData.
841	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
842	>	for (int i = 0; i < nLocal_; i++) {
843	>	// factor of two is because the total potential terms are divided
844	>	// by 2 in parallel due to row/ column scatter
845	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
846	>	}
847	>	}
848	>	}
849
850	+	if (storageLayout_ & DataStorage::dslParticlePot) {
851	+	int npp = snap_->atomData.particlePot.size();
852	+	vector<RealType> ppot_temp(npp, 0.0);
853	+
854	+	// This is the direct or embedding contribution to the particle
855	+	// pot.
856	+
857	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
858	+	for (int i = 0; i < npp; i++) {
859	+	snap_->atomData.particlePot[i] += ppot_temp[i];
860	+	}
861	+
862	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
863	+
864	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
865	+	for (int i = 0; i < npp; i++) {
866	+	snap_->atomData.particlePot[i] += ppot_temp[i];
867	+	}
868	+	}
869	+
870	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
871	+	RealType ploc1 = pairwisePot[ii];
872	+	RealType ploc2 = 0.0;
873	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
874	+	pairwisePot[ii] = ploc2;
875	+	}
876	+
877	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
878	+	RealType ploc1 = excludedPot[ii];
879	+	RealType ploc2 = 0.0;
880	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
881	+	excludedPot[ii] = ploc2;
882	+	}
883	+
884	+	// Here be dragons.
885	+	MPI::Intracomm col = colComm.getComm();
886	+
887	+	col.Allreduce(MPI::IN_PLACE,
888	+	&snap_->frameData.conductiveHeatFlux[0], 3,
889	+	MPI::REALTYPE, MPI::SUM);
890	+
891	+
892		#endif
893	+
894		}
895
896	+	/**
897	+	* Collects information obtained during the post-pair (and embedding
898	+	* functional) loops onto local data structures.
899	+	*/
900	+	void ForceMatrixDecomposition::collectSelfData() {
901	+	snap_ = sman_->getCurrentSnapshot();
902	+	storageLayout_ = sman_->getStorageLayout();
903	+
904	+	#ifdef IS_MPI
905	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
906	+	RealType ploc1 = embeddingPot[ii];
907	+	RealType ploc2 = 0.0;
908	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
909	+	embeddingPot[ii] = ploc2;
910	+	}
911	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
912	+	RealType ploc1 = excludedSelfPot[ii];
913	+	RealType ploc2 = 0.0;
914	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
915	+	excludedSelfPot[ii] = ploc2;
916	+	}
917	+	#endif
918	+
919	+	}
920	+
921	+
922	+
923		int ForceMatrixDecomposition::getNAtomsInRow() {
924		#ifdef IS_MPI
925		return nAtomsInRow_;
#	Line 659 | Line 960 | namespace OpenMD {
960		return d;
961		}
962
963	+	Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
964	+	#ifdef IS_MPI
965	+	return cgColData.velocity[cg2];
966	+	#else
967	+	return snap_->cgData.velocity[cg2];
968	+	#endif
969	+	}
970
971	+	Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
972	+	#ifdef IS_MPI
973	+	return atomColData.velocity[atom2];
974	+	#else
975	+	return snap_->atomData.velocity[atom2];
976	+	#endif
977	+	}
978	+
979	+
980		Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
981
982		Vector3d d;
#	Line 717 | Line 1034 | namespace OpenMD {
1034		return d;
1035		}
1036
1037	<	vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
1038	<	return skipsForAtom[atom1];
1037	>	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1038	>	return excludesForAtom[atom1];
1039		}
1040
1041		/**
1042	<	* There are a number of reasons to skip a pair or a
726	<	* particle. Mostly we do this to exclude atoms who are involved in
727	<	* short range interactions (bonds, bends, torsions), but we also
728	<	* need to exclude some overcounted interactions that result from
1042	>	* We need to exclude some overcounted interactions that result from
1043		* the parallel decomposition.
1044		*/
1045	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1046	<	int unique_id_1, unique_id_2;
1047	<
1045	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1046	>	int unique_id_1, unique_id_2, group1, group2;
1047	>
1048		#ifdef IS_MPI
1049		// in MPI, we have to look up the unique IDs for each atom
1050		unique_id_1 = AtomRowToGlobal[atom1];
1051		unique_id_2 = AtomColToGlobal[atom2];
1052	+	group1 = cgRowToGlobal[cg1];
1053	+	group2 = cgColToGlobal[cg2];
1054	+	#else
1055	+	unique_id_1 = AtomLocalToGlobal[atom1];
1056	+	unique_id_2 = AtomLocalToGlobal[atom2];
1057	+	group1 = cgLocalToGlobal[cg1];
1058	+	group2 = cgLocalToGlobal[cg2];
1059	+	#endif
1060
739	–	// this situation should only arise in MPI simulations
1061		if (unique_id_1 == unique_id_2) return true;
1062	<
1062	>
1063	>	#ifdef IS_MPI
1064		// this prevents us from doing the pair on multiple processors
1065		if (unique_id_1 < unique_id_2) {
1066		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1067		} else {
1068	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1068	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1069		}
1070	<	#else
1071	<	// in the normal loop, the atom numbers are unique
1072	<	unique_id_1 = atom1;
1073	<	unique_id_2 = atom2;
1070	>	#endif
1071	>
1072	>	#ifndef IS_MPI
1073	>	if (group1 == group2) {
1074	>	if (unique_id_1 < unique_id_2) return true;
1075	>	}
1076		#endif
1077
1078	<	for (vector<int>::iterator i = skipsForAtom[atom1].begin();
1079	<	i != skipsForAtom[atom1].end(); ++i) {
756	<	if ( (*i) == unique_id_2 ) return true;
757	<	}
1078	>	return false;
1079	>	}
1080
1081	+	/**
1082	+	* We need to handle the interactions for atoms who are involved in
1083	+	* the same rigid body as well as some short range interactions
1084	+	* (bonds, bends, torsions) differently from other interactions.
1085	+	* We'll still visit the pairwise routines, but with a flag that
1086	+	* tells those routines to exclude the pair from direct long range
1087	+	* interactions.  Some indirect interactions (notably reaction
1088	+	* field) must still be handled for these pairs.
1089	+	*/
1090	+	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1091	+
1092	+	// excludesForAtom was constructed to use row/column indices in the MPI
1093	+	// version, and to use local IDs in the non-MPI version:
1094	+
1095	+	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1096	+	i != excludesForAtom[atom1].end(); ++i) {
1097	+	if ( (*i) == atom2 ) return true;
1098	+	}
1099	+
1100	+	return false;
1101		}
1102
1103
#	Line 776 | Line 1118 | namespace OpenMD {
1118		}
1119
1120		// filling interaction blocks with pointers
1121	<	void ForceMatrixDecomposition::fillInteractionData(InteractionData idat,
1122	<	int atom1, int atom2) {
1121	>	void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1122	>	int atom1, int atom2) {
1123	>
1124	>	idat.excluded = excludeAtomPair(atom1, atom2);
1125	>
1126		#ifdef IS_MPI
1127	+	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1128	+	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1129	+	//                         ff_->getAtomType(identsCol[atom2]) );
1130
783	–	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
784	–	ff_->getAtomType(identsCol[atom2]) );
785	–
1131		if (storageLayout_ & DataStorage::dslAmat) {
1132		idat.A1 = &(atomRowData.aMat[atom1]);
1133		idat.A2 = &(atomColData.aMat[atom2]);
#	Line 818 | Line 1163 | namespace OpenMD {
1163		idat.particlePot2 = &(atomColData.particlePot[atom2]);
1164		}
1165
1166	<	#else
1166	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1167	>	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1168	>	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1169	>	}
1170
1171	<	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
1172	<	ff_->getAtomType(identsLocal[atom2]) );
1171	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1172	>	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1173	>	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1174	>	}
1175
1176	+	#else
1177	+
1178	+	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1179	+
1180		if (storageLayout_ & DataStorage::dslAmat) {
1181		idat.A1 = &(snap_->atomData.aMat[atom1]);
1182		idat.A2 = &(snap_->atomData.aMat[atom2]);
#	Line 838 | Line 1192 | namespace OpenMD {
1192		idat.t2 = &(snap_->atomData.torque[atom2]);
1193		}
1194
1195	<	if (storageLayout_ & DataStorage::dslDensity) {
1195	>	if (storageLayout_ & DataStorage::dslDensity) {
1196		idat.rho1 = &(snap_->atomData.density[atom1]);
1197		idat.rho2 = &(snap_->atomData.density[atom2]);
1198		}
#	Line 858 | Line 1212 | namespace OpenMD {
1212		idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1213		}
1214
1215	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1216	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1217	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1218	+	}
1219	+
1220	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1221	+	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1222	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1223	+	}
1224	+
1225		#endif
1226		}
1227
1228
1229	<	void ForceMatrixDecomposition::unpackInteractionData(InteractionData idat, int atom1, int atom2) {
1229	>	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1230		#ifdef IS_MPI
1231	<	pot_row[atom1] += 0.5 *  *(idat.pot);
1232	<	pot_col[atom2] += 0.5 *  *(idat.pot);
1231	>	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1232	>	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1233	>	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1234	>	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1235
1236		atomRowData.force[atom1] += *(idat.f1);
1237		atomColData.force[atom2] -= *(idat.f1);
872	–	#else
873	–	longRangePot_ += *(idat.pot);
874	–
875	–	snap_->atomData.force[atom1] += *(idat.f1);
876	–	snap_->atomData.force[atom2] -= *(idat.f1);
877	–	#endif
1238
1239	<	}
1239	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
1240	>	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1241	>	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1242	>	}
1243
1244	+	if (storageLayout_ & DataStorage::dslElectricField) {
1245	+	atomRowData.electricField[atom1] += *(idat.eField1);
1246	+	atomColData.electricField[atom2] += *(idat.eField2);
1247	+	}
1248
1249	<	void ForceMatrixDecomposition::fillSkipData(InteractionData idat,
1250	<	int atom1, int atom2) {
1251	<	#ifdef IS_MPI
885	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
886	<	ff_->getAtomType(identsCol[atom2]) );
1249	>	#else
1250	>	pairwisePot += *(idat.pot);
1251	>	excludedPot += *(idat.excludedPot);
1252
1253	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1254	<	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
1255	<	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1253	>	snap_->atomData.force[atom1] += *(idat.f1);
1254	>	snap_->atomData.force[atom2] -= *(idat.f1);
1255	>
1256	>	if (idat.doParticlePot) {
1257	>	// This is the pairwise contribution to the particle pot.  The
1258	>	// embedding contribution is added in each of the low level
1259	>	// non-bonded routines.  In parallel, this calculation is done
1260	>	// in collectData, not in unpackInteractionData.
1261	>	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1262	>	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1263		}
1264	<	if (storageLayout_ & DataStorage::dslTorque) {
1265	<	idat.t1 = &(atomRowData.torque[atom1]);
1266	<	idat.t2 = &(atomColData.torque[atom2]);
1264	>
1265	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
1266	>	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1267	>	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1268		}
896	–	#else
897	–	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
898	–	ff_->getAtomType(identsLocal[atom2]) );
1269
1270	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1271	<	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1272	<	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1270	>	if (storageLayout_ & DataStorage::dslElectricField) {
1271	>	snap_->atomData.electricField[atom1] += *(idat.eField1);
1272	>	snap_->atomData.electricField[atom2] += *(idat.eField2);
1273		}
1274	<	if (storageLayout_ & DataStorage::dslTorque) {
1275	<	idat.t1 = &(snap_->atomData.torque[atom1]);
1276	<	idat.t2 = &(snap_->atomData.torque[atom2]);
907	<	}
908	<	#endif
1274	>
1275	>	#endif
1276	>
1277		}
1278
1279		/*
#	Line 918 | Line 1286 | namespace OpenMD {
1286
1287		vector<pair<int, int> > neighborList;
1288		groupCutoffs cuts;
1289	+	bool doAllPairs = false;
1290	+
1291		#ifdef IS_MPI
1292		cellListRow_.clear();
1293		cellListCol_.clear();
#	Line 937 | Line 1307 | namespace OpenMD {
1307		nCells_.y() = (int) ( Hy.length() )/ rList_;
1308		nCells_.z() = (int) ( Hz.length() )/ rList_;
1309
1310	+	// handle small boxes where the cell offsets can end up repeating cells
1311	+
1312	+	if (nCells_.x() < 3) doAllPairs = true;
1313	+	if (nCells_.y() < 3) doAllPairs = true;
1314	+	if (nCells_.z() < 3) doAllPairs = true;
1315	+
1316		Mat3x3d invHmat = snap_->getInvHmat();
1317		Vector3d rs, scaled, dr;
1318		Vector3i whichCell;
1319		int cellIndex;
1320		int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1321
946	–	cerr << "flag1\n";
1322		#ifdef IS_MPI
1323		cellListRow_.resize(nCtot);
1324		cellListCol_.resize(nCtot);
1325		#else
1326		cellList_.resize(nCtot);
1327		#endif
1328	<	cerr << "flag2\n";
1328	>
1329	>	if (!doAllPairs) {
1330		#ifdef IS_MPI
955	–	for (int i = 0; i < nGroupsInRow_; i++) {
956	–	rs = cgRowData.position[i];
1331
1332	<	// scaled positions relative to the box vectors
1333	<	scaled = invHmat * rs;
1334	<
1335	<	// wrap the vector back into the unit box by subtracting integer box
1336	<	// numbers
1337	<	for (int j = 0; j < 3; j++) {
1338	<	scaled[j] -= roundMe(scaled[j]);
1339	<	scaled[j] += 0.5;
1332	>	for (int i = 0; i < nGroupsInRow_; i++) {
1333	>	rs = cgRowData.position[i];
1334	>
1335	>	// scaled positions relative to the box vectors
1336	>	scaled = invHmat * rs;
1337	>
1338	>	// wrap the vector back into the unit box by subtracting integer box
1339	>	// numbers
1340	>	for (int j = 0; j < 3; j++) {
1341	>	scaled[j] -= roundMe(scaled[j]);
1342	>	scaled[j] += 0.5;
1343	>	}
1344	>
1345	>	// find xyz-indices of cell that cutoffGroup is in.
1346	>	whichCell.x() = nCells_.x() * scaled.x();
1347	>	whichCell.y() = nCells_.y() * scaled.y();
1348	>	whichCell.z() = nCells_.z() * scaled.z();
1349	>
1350	>	// find single index of this cell:
1351	>	cellIndex = Vlinear(whichCell, nCells_);
1352	>
1353	>	// add this cutoff group to the list of groups in this cell;
1354	>	cellListRow_[cellIndex].push_back(i);
1355		}
1356	<
1357	<	// find xyz-indices of cell that cutoffGroup is in.
1358	<	whichCell.x() = nCells_.x() * scaled.x();
1359	<	whichCell.y() = nCells_.y() * scaled.y();
1360	<	whichCell.z() = nCells_.z() * scaled.z();
1361	<
1362	<	// find single index of this cell:
1363	<	cellIndex = Vlinear(whichCell, nCells_);
1364	<
1365	<	// add this cutoff group to the list of groups in this cell;
1366	<	cellListRow_[cellIndex].push_back(i);
1367	<	}
1368	<
1369	<	for (int i = 0; i < nGroupsInCol_; i++) {
1370	<	rs = cgColData.position[i];
1371	<
1372	<	// scaled positions relative to the box vectors
1373	<	scaled = invHmat * rs;
1374	<
1375	<	// wrap the vector back into the unit box by subtracting integer box
1376	<	// numbers
1377	<	for (int j = 0; j < 3; j++) {
1378	<	scaled[j] -= roundMe(scaled[j]);
990	<	scaled[j] += 0.5;
1356	>	for (int i = 0; i < nGroupsInCol_; i++) {
1357	>	rs = cgColData.position[i];
1358	>
1359	>	// scaled positions relative to the box vectors
1360	>	scaled = invHmat * rs;
1361	>
1362	>	// wrap the vector back into the unit box by subtracting integer box
1363	>	// numbers
1364	>	for (int j = 0; j < 3; j++) {
1365	>	scaled[j] -= roundMe(scaled[j]);
1366	>	scaled[j] += 0.5;
1367	>	}
1368	>
1369	>	// find xyz-indices of cell that cutoffGroup is in.
1370	>	whichCell.x() = nCells_.x() * scaled.x();
1371	>	whichCell.y() = nCells_.y() * scaled.y();
1372	>	whichCell.z() = nCells_.z() * scaled.z();
1373	>
1374	>	// find single index of this cell:
1375	>	cellIndex = Vlinear(whichCell, nCells_);
1376	>
1377	>	// add this cutoff group to the list of groups in this cell;
1378	>	cellListCol_[cellIndex].push_back(i);
1379		}
1380	<
993	<	// find xyz-indices of cell that cutoffGroup is in.
994	<	whichCell.x() = nCells_.x() * scaled.x();
995	<	whichCell.y() = nCells_.y() * scaled.y();
996	<	whichCell.z() = nCells_.z() * scaled.z();
997	<
998	<	// find single index of this cell:
999	<	cellIndex = Vlinear(whichCell, nCells_);
1000	<
1001	<	// add this cutoff group to the list of groups in this cell;
1002	<	cellListCol_[cellIndex].push_back(i);
1003	<	}
1380	>
1381		#else
1382	<	for (int i = 0; i < nGroups_; i++) {
1383	<	rs = snap_->cgData.position[i];
1384	<
1385	<	// scaled positions relative to the box vectors
1386	<	scaled = invHmat * rs;
1387	<
1388	<	// wrap the vector back into the unit box by subtracting integer box
1389	<	// numbers
1390	<	for (int j = 0; j < 3; j++) {
1391	<	scaled[j] -= roundMe(scaled[j]);
1392	<	scaled[j] += 0.5;
1382	>	for (int i = 0; i < nGroups_; i++) {
1383	>	rs = snap_->cgData.position[i];
1384	>
1385	>	// scaled positions relative to the box vectors
1386	>	scaled = invHmat * rs;
1387	>
1388	>	// wrap the vector back into the unit box by subtracting integer box
1389	>	// numbers
1390	>	for (int j = 0; j < 3; j++) {
1391	>	scaled[j] -= roundMe(scaled[j]);
1392	>	scaled[j] += 0.5;
1393	>	}
1394	>
1395	>	// find xyz-indices of cell that cutoffGroup is in.
1396	>	whichCell.x() = nCells_.x() * scaled.x();
1397	>	whichCell.y() = nCells_.y() * scaled.y();
1398	>	whichCell.z() = nCells_.z() * scaled.z();
1399	>
1400	>	// find single index of this cell:
1401	>	cellIndex = Vlinear(whichCell, nCells_);
1402	>
1403	>	// add this cutoff group to the list of groups in this cell;
1404	>	cellList_[cellIndex].push_back(i);
1405		}
1406
1018	–	// find xyz-indices of cell that cutoffGroup is in.
1019	–	whichCell.x() = nCells_.x() * scaled.x();
1020	–	whichCell.y() = nCells_.y() * scaled.y();
1021	–	whichCell.z() = nCells_.z() * scaled.z();
1022	–
1023	–	// find single index of this cell:
1024	–	cellIndex = Vlinear(whichCell, nCells_);
1025	–
1026	–	// add this cutoff group to the list of groups in this cell;
1027	–	cellList_[cellIndex].push_back(i);
1028	–	}
1407		#endif
1408
1409	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1410	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1411	<	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1412	<	Vector3i m1v(m1x, m1y, m1z);
1413	<	int m1 = Vlinear(m1v, nCells_);
1036	<
1037	<	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1038	<	os != cellOffsets_.end(); ++os) {
1409	>	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1410	>	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1411	>	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1412	>	Vector3i m1v(m1x, m1y, m1z);
1413	>	int m1 = Vlinear(m1v, nCells_);
1414
1415	<	Vector3i m2v = m1v + (*os);
1416	<
1417	<	if (m2v.x() >= nCells_.x()) {
1418	<	m2v.x() = 0;
1419	<	} else if (m2v.x() < 0) {
1045	<	m2v.x() = nCells_.x() - 1;
1046	<	}
1047	<
1048	<	if (m2v.y() >= nCells_.y()) {
1049	<	m2v.y() = 0;
1050	<	} else if (m2v.y() < 0) {
1051	<	m2v.y() = nCells_.y() - 1;
1052	<	}
1053	<
1054	<	if (m2v.z() >= nCells_.z()) {
1055	<	m2v.z() = 0;
1056	<	} else if (m2v.z() < 0) {
1057	<	m2v.z() = nCells_.z() - 1;
1058	<	}
1059	<
1060	<	int m2 = Vlinear (m2v, nCells_);
1415	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1416	>	os != cellOffsets_.end(); ++os) {
1417	>
1418	>	Vector3i m2v = m1v + (*os);
1419	>
1420
1421	<	#ifdef IS_MPI
1422	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1423	<	j1 != cellListRow_[m1].end(); ++j1) {
1424	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1425	<	j2 != cellListCol_[m2].end(); ++j2) {
1426	<
1427	<	// Always do this if we're in different cells or if
1428	<	// we're in the same cell and the global index of the
1429	<	// j2 cutoff group is less than the j1 cutoff group
1421	>	if (m2v.x() >= nCells_.x()) {
1422	>	m2v.x() = 0;
1423	>	} else if (m2v.x() < 0) {
1424	>	m2v.x() = nCells_.x() - 1;
1425	>	}
1426	>
1427	>	if (m2v.y() >= nCells_.y()) {
1428	>	m2v.y() = 0;
1429	>	} else if (m2v.y() < 0) {
1430	>	m2v.y() = nCells_.y() - 1;
1431	>	}
1432	>
1433	>	if (m2v.z() >= nCells_.z()) {
1434	>	m2v.z() = 0;
1435	>	} else if (m2v.z() < 0) {
1436	>	m2v.z() = nCells_.z() - 1;
1437	>	}
1438
1439	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1439	>	int m2 = Vlinear (m2v, nCells_);
1440	>
1441	>	#ifdef IS_MPI
1442	>	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1443	>	j1 != cellListRow_[m1].end(); ++j1) {
1444	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1445	>	j2 != cellListCol_[m2].end(); ++j2) {
1446	>
1447	>	// In parallel, we need to visit *all* pairs of row
1448	>	// & column indicies and will divide labor in the
1449	>	// force evaluation later.
1450		dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1451		snap_->wrapVector(dr);
1452		cuts = getGroupCutoffs( (*j1), (*j2) );
1453		if (dr.lengthSquare() < cuts.third) {
1454		neighborList.push_back(make_pair((*j1), (*j2)));
1455	<	}
1455	>	}
1456		}
1457		}
1081	–	}
1458		#else
1459	+	for (vector<int>::iterator j1 = cellList_[m1].begin();
1460	+	j1 != cellList_[m1].end(); ++j1) {
1461	+	for (vector<int>::iterator j2 = cellList_[m2].begin();
1462	+	j2 != cellList_[m2].end(); ++j2) {
1463	+
1464	+	// Always do this if we're in different cells or if
1465	+	// we're in the same cell and the global index of
1466	+	// the j2 cutoff group is greater than or equal to
1467	+	// the j1 cutoff group.  Note that Rappaport's code
1468	+	// has a "less than" conditional here, but that
1469	+	// deals with atom-by-atom computation.  OpenMD
1470	+	// allows atoms within a single cutoff group to
1471	+	// interact with each other.
1472
1084	–	for (vector<int>::iterator j1 = cellList_[m1].begin();
1085	–	j1 != cellList_[m1].end(); ++j1) {
1086	–	for (vector<int>::iterator j2 = cellList_[m2].begin();
1087	–	j2 != cellList_[m2].end(); ++j2) {
1473
1089	–	// Always do this if we're in different cells or if
1090	–	// we're in the same cell and the global index of the
1091	–	// j2 cutoff group is less than the j1 cutoff group
1474
1475	<	if (m2 != m1 || (*j2) < (*j1)) {
1476	<	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1477	<	snap_->wrapVector(dr);
1478	<	cuts = getGroupCutoffs( (*j1), (*j2) );
1479	<	if (dr.lengthSquare() < cuts.third) {
1480	<	neighborList.push_back(make_pair((*j1), (*j2)));
1475	>	if (m2 != m1 || (*j2) >= (*j1) ) {
1476	>
1477	>	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1478	>	snap_->wrapVector(dr);
1479	>	cuts = getGroupCutoffs( (*j1), (*j2) );
1480	>	if (dr.lengthSquare() < cuts.third) {
1481	>	neighborList.push_back(make_pair((*j1), (*j2)));
1482	>	}
1483		}
1484		}
1485		}
1102	–	}
1486		#endif
1487	+	}
1488		}
1489		}
1490		}
1491	+	} else {
1492	+	// branch to do all cutoff group pairs
1493	+	#ifdef IS_MPI
1494	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1495	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1496	+	dr = cgColData.position[j2] - cgRowData.position[j1];
1497	+	snap_->wrapVector(dr);
1498	+	cuts = getGroupCutoffs( j1, j2 );
1499	+	if (dr.lengthSquare() < cuts.third) {
1500	+	neighborList.push_back(make_pair(j1, j2));
1501	+	}
1502	+	}
1503	+	}
1504	+	#else
1505	+	// include all groups here.
1506	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1507	+	// include self group interactions j2 == j1
1508	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1509	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1510	+	snap_->wrapVector(dr);
1511	+	cuts = getGroupCutoffs( j1, j2 );
1512	+	if (dr.lengthSquare() < cuts.third) {
1513	+	neighborList.push_back(make_pair(j1, j2));
1514	+	}
1515	+	}
1516	+	}
1517	+	#endif
1518		}
1519	<
1519	>
1520		// save the local cutoff group positions for the check that is
1521		// done on each loop:
1522		saved_CG_positions_.clear();

Diff Legend

–	Removed lines
+	Added lines
<	Changed lines
>	Changed lines